[CPMD-list] Energy components in PIMD

[Axel Kohlmeyer]

On Thu, 8 May 2008, ukasz Walewski wrote:

LW> Dnia 2008-04-29, wto o godzinie 07:26 -0400, Axel Kohlmeyer pisze:
LW> > 
LW> > dear lukasz,
LW> > 
LW> > can your input conserve energy for a non-PIMD run?
LW> > ...and then you have to check whether you can maintain
LW> > adiabacity with the massive thermostat on the ions only.
LW> > quite a few times, the no-thermostat CP-dynamics is only
LW> > meta-stable and when you switch to massive N-H chains you
LW> > excite DOFs that will make the CP dynamics lose it.
LW> 
LW> Going even further isn't it that the electrons in NVT simulation tend to
LW> achieve the energy equipartition, i.e. the kinetic energy of the
LW> electronic DOFs rises to achieve the level of 0.5kT per DOF ? And since
LW> we allow the energy gain from the ionic thermostat the electrons just
LW> absorb it. This would mean that there is a strong coupling between the
LW> electronic and ionic DOF in my simulation which is exactly the CPMD
LW> catastrophe. In this case decreasing the electron mass should improve
LW> this.

in your specific case, the situation you have to watch out for is
the transfer of protons (i guess that is why you want to run 
path-integrals in the first place). if the vibrational DOF that
would loosen a proton from its nitrogen is not yet excited, it
should be excited pretty fast with massive NH chains. that in turn
may require that you in fact reduce the electron mass and the 
time step so that you stay close to the BO surface _and_ maintain
adiabacity. of course you can also "stabilize" the CP-dynamics by
increasing the fictitious mass, but that would also increase the 
"drag", i.e. the effective mass of the ions and that would 
ultimately modify the barrier height for the proton transfer.

another situation that can happen, but is _less_ likely for your
specific system is that you start from an (idealized) structure 
and the initial wavefunction optimization converges to the wrong
state. now you run your CP dynamics and (initially) symmetries are
preserved and nothing bad happens, then you switch to massice NH chains
and suddenly the electronic system can leave the local minimum it
was trapped and you get a large increase of fictitious energy.
this is most likely to happen, when the intial atomic guess is
quite different from the final structure for a few elements (e.g.
transition metals) and in that case it is in many cases better
to start the wavefunction optimization from a random wavefunction
and/or use CP dynamics with annealing and all atom positions fixed
(BTW: it would be great is somebody implemented some damped dynamics
scheme for that which is quite efficient and rather simple to do).

cheers,
   axel.
LW> 
LW> Could you please comment on that, is there a seed of reason or did I
LW> lost the point totally ?
LW> 
LW> Regards,
LW> Lukasz
LW> 

-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.